Advances in processor technology have made it possible to use general-purpose computers as real-time digital signal processors in musical composition and multimedia applications. Real-time software synthesis and audio coding algorithms execute differently on different host hardware and operating systems. Our work explores the suitability of scalable sound representations for heterogeneous hosts and networks. The model we have developed is based on asynchronous granular synthesis. Two different representations are included in our model; an unbound granular model and a traditional grid-like structure. The first model allows the most flexibility for the composer, an almost unbound range of granular data of unlimited density can be used, and scheduling in the synthesiser decides which grains should be used at any one time. This is normally dependant on available computation time and the quality of output required by the application. The grid model was developed so that we could apply a similar level of real-time adaptation to natural sound signals. This paper concentrates on the design of a real-time asynchronous granular synthesis system, with scalability achieved through both real-time scheduling and continuously varying rendering quality. Experimental results are presented which show the behaviour of the synthesis system across a variety of hosts. Also presented is an analysis of how the extensive optimisation affects the audio quality in the synthesis system.
展开▼